Have used a few coding models and they are good, but they will not replace a programmer. You can ask for a class to be made, or a method. You can ask it to find issues or improvements with/for your code or spot mistakes, but it can’t hack multiple modules and give you a fully working complex app. It’ll help you learn to code, it’ll write you whatever you ask, but when you start adding the fact you need to know about X for y and z is also important it starts forgetting the original prompt.
Don’t get me wrong. It’s really good. But it’s a tool, not a full on master coder.
Looking forward to getting more tokens to work with on a normal computer. Happy to wait for the CPU to gen code considering the ridiculous prices of high VRAM GPUs these days, but it’s still fisher price ‘My first AI coding assistant’. Give it a few more years, a few more breakthroughs and we will get there.
I have tried GPT Engineer… it’s not there yet. Will make you a simple app, but it’s not going to knock out anything more than modular microshite.
I’ve found they really help with unit tests. Sometimes with regular code they straight up make up libraries that aren’t real.
Officially, this time 😂
Is this article trying to imply meta had a choice in the matter when llama’s code has been available online for the better part of a year?
And if it was actually good and fully usable they’d just use it themselves to make billions…
Imagine you could just write a specification for any software project in the world (even invest some programmers and project managers to do it together with reviews), throw it at the AI and you get a fully functioning program out. Preferably even with a DevOps pipeline set up, so all you have to do is copy it into AWS, Azure or whatever, throw money at it and you’re done.
They wouldn’t give that away, lol.
Definitely not how it works. It’ll be a tool for programmers to make them more efficient. It’ll be like the difference between a hammer and nails and a nail gun. You still need the worker to know where to put the nail.
Writing code is fast, checking it’s correct and debugging is slow. If some AI spits out 2000 lines of code it won’t make you faster at all.
Have you tried asking ChatGPT or Bard to write you code to do something? It is actually remarkably good at it.
That and being an alternative to a thesaurus is about all I use LLMs for.Lol, yeah, and it hallucinates all the time. You also use it to just write a little bit of new code, you can’t give it a 100k lines code base and tell it to actually add or modify something…
Indeed, and for me giving me a rough framework to modify is hugely useful and time saving. As the commenter above said it’s a tool, it’s not a team member.
Also many programmers write their own stuff for generating boilerplate code, with some languages like D having a strong support for it via its metaprogramming features. All while being more reliable.
How does one use it?
